Second-class Intel to trail AMD for years

[endyen]

that's just your opinion that it's corporate trash

Hmm, I called it "corporate line", and it really doesn't take much to recognize an Intel (or nvidia, or Amd, or Ati) phrase when it appears.

The only major missing difference is the chip-to-chip HTT, which Intel has their own take on anyway.

Not that I know of. That's the problem. Intel has a crosstalk, for core to core on one chip, but chip to chip, they will use the fsb.
No matter how you look at it, scotty still has two major problems. If SOI can take care of the heat problem, a quad pumped 400mhz fsb, with sync pc6400 should negate the bandwidth problem. Even @ 3.8ghz, that would yield impressive numbers.

 

[ltcommander_data]

I think the major mistake that Intel made with the Prescott architecture was increasing the pipeline to 31 stages. If it wasn't for the lengthened pipeline and its associated heat and power issues Netburst would probably be alive and kicking today.

Excluding the pipeline lengthening, the improvements made to Prescott were pretty impressive. The improvements in branch prediction and the scheduler helped improvement effiency within the processor. While the processing power was also improved with the addition of a shift/rotater to a double-pumped fast ALU. Similarly, the addition of a dedicated integer multiplier freed up the FPU for other tasks. What's more the L1 cache was doubled to 16kb with associativity increased to 8-way. The trace cache has also been improved and the L2 cache quadrupled to 2MB in the Prescott 2M. Hyperthreading was also improved.

Sadly, all these improvements were used to hide the lengthened pipeline in Prescott. Which, while it didn't do completely, was fairly effective against the 55% increase in the pipelline. If Intel had kept the pipeline of Prescott the same length as Northwood while implementing all of Prescott's other improvements Netburst would still be competitive. Intel should have just relied on the shift from 130nm to 90nm to bring increases in clock speeds. Granted the clock speed increases allowed by relying just on a process shrink wouldn't be as great as the lengthening of the pipeline, combined with the other improvements, a Northwood with Prescott improvements would have remained competitive with AMD.

It is curious why Intel doesn't implement SOI. Even without SOI though, what Netburst needs to save it is Intel's upcoming 45nm process. Supposedly Intel has solved leakage problems with the 45nm process allowing phenominal power and heat reductions.

The Inquirer seems very excited by the prospects of the new process:

"So, if you hear gushingly good things about 45nm coming from IDF, believe it. If you hear anyone pooh-poohing Intel and its process tech because of the debacle that was 90nm, just point and laugh. This one will be very very good."

Sadly, Netburst won't survive long enough to realize the benefits of the 45nm process. However, I can only imagine the gains in clock speed, power reduction and heat production on Intel's already efficient next-generation Conroe, Meron, and Woodcrest architecture.

The Inquirer saids:

"Think happy thoughts here people, from what several sources have told the INQ, the leakage problem is solved, and I mean solved, not lessened. This will be a massive gain for Intel, and unless AMD and IBM can match it, it will pretty much hand it the mobile space, not to mention anything else where power matters."

Very promising.

http://www.theinquirer.net/?article=25512

 

[ltcommander_data]

I found a new article.

http://www.theinquirer.net/?article=27421

It seems that AMD's next-generation K10 architecture is dead. This is actually kind of funny really since eveeryone has been pointing fingers at Intel for delaying the introduction of Xeons with integrated memory controllers. Now it seems AMD is facing problems as well.

Without the K10, the AMD64 will definitely lag behind Intel's next-generation Conroe, Meron, and Woodcrest architecture. Already a Dothan at 2.56GHz Dothan is able to beat AMD's FX-55 and Intel's own Pentium 4 Extreme Editions in all 3d games as well as in most other benchmarks. It also did so while running at 58% lower power comsumption and the FX-55. And this is only a Dothan. Yonah itself will offer considerable improvement over Dothan with its improvements to multimedia and FPU performance. One can only imagine Conroe's potential.

To meet Intel's new architecture, AMD will only have a stop gap measure called the K8L. With the K10 or its replacement delayed until 2008, Intel will have an excellent opportunity to dominate AMD and regain market share over the next 3 years. Hopefully, the execs at Intel perk up and don't squander it.

The Dothan figures are from this article:
http://www.tomshardware.com/cpu/20050525/pentium4-21.html

I felt this probably deserved its own thread so I created one. I'm just leaving this here as reference.<P ID="edit"><FONT SIZE=-1><EM>Edited by ltcommander_data on 11/03/05 10:20 PM.</EM></FONT></P>

 

[endyen]

You assume that K10 is having problems. AFAIK Amd has faster desktop chips ready now, but they aren't releasing them. I suspect that Amd is afraid of getting too far ahead of Intel. They can probably bring K10 out just about any time now, but why bother. They already have a clear lead.
I just thought of something. The new dual cores will require a new motherboard. Had I bought a xeon system in the last 6 months to a year, I would be pissed. Now, Intel is planning on a new platform for their next batch of xeons as well. If you have a xeon now, you can't upgrade to dual cores. If you buy a dual core setup, it will only be supported for about a year. Not a good way to deal with business customers is it?

 

[Xeon]

Dude I would simply walk away from the thread this is the THG infinite loop all the morons come in waves spewing the same same hate anti intel and anti you babble.

If you want to have a real intellectual conversation with someone about technology this isnt the place to be the kiddies and trolls come out and it becomes impossible to discuss anything.

-Jeremy Dach

 

[slvr_phoenix]

Um, no.

The major missing difference is that the Intel solution still leaves all cores competing for memory bandwidth. Four FSBs isn't enough to compete with HTT unless each FSB gets its own memory bank

If you'd been following along, that is exactly the plan, at least as far as I've seen. Quad-channel DDR2. Four FSBs. A perfect extension of Intel's current working methods to overcome the most notable problems. Silly? Yep. Expensive? Definately. Competetive? Should do.

Never mind the extra latency Intel's solution introduces for chip-to-chip traffic.

Exactly why I've never said that Intel's method is better. Because it <i>isn't</i> better. It's just <i>almost</i> as good. Barring major cockups it should be able to <i>compete</i>. Not win, but compete.


یί∫υєг ρђœŋίχ
The <b><font color=red>Devil</font color=red></b> is in my <b><font color=red>'98 Mercury Sable</font color=red></b>!
<b>@ 201K miles!</b>

 

[slvr_phoenix]

Hmm, I called it "corporate line", and it really doesn't take much to recognize an Intel (or nvidia, or Amd, or Ati) phrase when it appears.

1) You <i>called</i> it "corporate line", but your tone and demeanor make it clear the value that you give that line.
2) Why is it so wrong for anyone to personally believe that "corporate line"?

Not that I know of. That's the problem. Intel has a crosstalk, for core to core on one chip, but chip to chip, they will use the fsb.

**taps mic** Hello? Is this thing on? That's exactly what I said. Intel has their own take on it. A 4xFSB and crosstalk. It's certainly no HTT, but it is still most definately <i>Intel's take on it</i>. Clearly Intel has a different point of view than AMD on this.

No matter how you look at it, scotty still has two major problems.

Scotty will <i>always</i> have one major problem, no matter what else Intel fixes, because it's a part of the core itself. And that's the way Intel dumped a lot more latency into the L2 cache (so that the core would clock higher) and had to screw up what had been a smooth mispredict in Northy to accomodate that massive addition to latency in Scotty. Even if Intel fixes everything else, Scotty will still suck IMHO for that reason alone. Scotty <i>could</i> have been a Northy killer. Instead, for all of it's improvements, it's still not even able to compete on the same scale, and that's a pretty major problem.

If SOI can take care of the heat problem, a quad pumped 400mhz fsb, with sync pc6400 should negate the bandwidth problem. Even @ 3.8ghz, that would yield impressive numbers.

It is impressive enough to be competetive anyway. I doubt that Intel will ever manage to take AMD again until they really fix their core problems, which funnily enough, are both latency, one in the L2, and one in the memory controller on the NB instead of OD.

Methinks that Netburst will fail badly if Intel doesn't re-examine their original plans for it and get back to them. Scotty deviated too far, and for all the wrong reasons.


یί∫υєг ρђœŋίχ
The <b><font color=red>Devil</font color=red></b> is in my <b><font color=red>'98 Mercury Sable</font color=red></b>!
<b>@ 201K miles!</b>

 

[Era]

Yea, I think Xeon right, again.
We get all kinds of posts from experts in here.
Just look at the one below. I wonder who wrote it?

<b>~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
K ill try and not post BS just for you, simply because you asked.



In reply to:
--------------------------------------------------------------------------------

*note, I know it aint gonna happen, so keep up the good work. Your seperate reality is always a pleasure, Spuddy ole buddy.

--------------------------------------------------------------------------------

Ha ha oh my did you think that up or did your wife help? That sir is the humor I enjoy snide yet civilized *tips hat*.

But hey choose not to read what I say I disregard everything you guys say since your all closet cases.

*snickers*
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
</b>

 

[HansGruber]

Netburst is the worst mistake Intel has made in history EVER!

It was excellent business decision.
Intel did manage to get good sales with "higher MHZ is better" myth, since people usually believe that bigger the number better it is.

<font color=red>"We can't solve problems by using the same kind of thinking we used when we created them."
- Albert Einstein</font color=red>

 

[Kelledin]

If you'd been following along, that is exactly the plan, at least as far as I've seen. Quad-channel DDR2. Four FSBs. A perfect extension of Intel's current working methods to overcome the most notable problems. Silly? Yep. Expensive? Definately. Competetive? Should do.

Oh, it goes beyond being silly and expensive. It's absurdly non-feasible for all but the most expensive x86 servers. A memory solution like that might well require something along the lines of a 12-layer motherboard PCB. People were already complaining about Opterons taking 6-layer boards and recent Radeons taking something like 9-layer PCB. In fact, the only places i've seen 12-layer PCB are in hot-swap PSU combiner boards carrying up to 1200W, Slot B EV6x CPU modules, and possibly the UP2000 boards from API Networks. All were horrendously expensive (and the combiner boards had to be beefed up even more to pass Hi-POT).

Basically this solution is only good for the price-no-object, absolutely-must-have-Intel situations, like the recent IBM Paxville benchmark that took a much larger disk array, a higher-end database, and a tripled price-tag to squeak by the Opteron competition in TPCc. At that price (not to mention a larger volumetric and heat footprint), it had better do a hell of a lot more than squeak by a rival from seven months past.

"You have been sh<font color=black>it</font color=black> upon by a grue."

 

[zeezee]

It was excellent business decision.

Short term (and short sighted), Yes.

 

[Xeon]

The short version is Intel fu<b></b>cked up, but not totally thanx to Northwood.
It it was something like Willy to Scotty with no Northwood-ish in between then Intel would've trailed AMD even further.

AMD sold 52% of processors last month. End of story.

<b><font color=red>Prescott=FAILURE</font color=red>
ASUS+AMD=Prescott</font color=green></b>

And what does it matter that Intel will globally sell 80+/-% of the markets silicon consumption.

 

[K8MAN]

The short version is Intel fu<b></b>cked up, but not totally thanx to Northwood.
It it was something like Willy to Scotty with no Northwood-ish in between then Intel would've trailed AMD even further.

AMD sold 52% of processors last month. End of story.

<b><font color=red>Prescott=FAILURE</font color=red>
ASUS+AMD=Prescott</font color=green></b>

And what does it matter that Intel will globally sell 80+/-% of the markets silicon consumption.
Ford sell's a lot of cars but i'd stil rather drive a BMW.

 

[K8MAN]

And btw u should come work @ Dell with me. Its so easy its not even a job and its cheaper to live in Edmonton than Lloyd.

 

[Rob423]

ha, like said above, they deff made there money... I don't think the "Naming" scheme had much of a problem... i think they just released way to many cpus instead of releasing a fewer amount and just making that product better.

Oh well... It's a good time for AMD customers!

 

[AmdMELTDOWN]

lol, do I need to read this whole thread, when it's all in my sig.

 

[rettihSlluB]

Motherboard makers: Intel’s dual-core Presler has stability issues

Once again all-mighty Intel going nasty. :roll:

 

[slvr_phoenix]

Motherboard makers: Intel’s dual-core Presler has stability issues

Once again all-mighty Intel going nasty. :roll:

:lol: What do you think engineering samples are for if not to find the bugs? :mrgreen:

I'd be more scared if they didn't find them during the testing phase.

 

[ltcommander_data]

http://www.xbitlabs.com/news/cpu/display/20051115161111.html

"It is reported that only lower-clocked versions are affected."

The problem doesn't appear that bad though since its isolated to the lower-clocked versions, which indicates it isn't a fundamental design flaw in Presler. Since the motherboard makers themselves reported it, the problem might be similar to the stability problems the nVidia nForce4 SLI has with the lower clocked Smithfield 820.

What gets me more excited is the possibility that Yonah will actually be launched this year instead of in January.

http://www.theregister.com/2005/11/15/intel_presler_problems/

"Meanwhile, 'Yonah', the 65nm dual-core Pentium M chip, has apparently passed the appropriate tests, the sources claim, and is on track to ship later this year, just ahead of its formal introduction early next year."

A December launch for Yonah seems likely in order to put it ahead of the MacWorld that Steve Jobs is planning in early January to launch the new Intel Macs which are based on Yonah. They could also combine the launch with the new Celeron speed bump that is planned for December 25. Might as well make an early launch of the 955EE as well, like what the 840EE got since it doesn't appear to have any stability problems. Seeing a 3.46GHz Dempsey competes favourably against the Opteron 280, its probably in Intel's best interest to ship the 955EE ASAP. Since the FX-60 is due to ship early January and appears faster, Intel would need to ship the 955EE while it can still compete with the X2 4800+.

Personally, I'm waiting for Yonah before I get a new labtop so the sooner the better. Hopefully before Christmas.

 

[endyen]

Well now I am a little confused.
Almost a month ago, Intel announced that they had begun volumn shipments of preslers. Now we hear that board makers are still working with engineering samples. Dont they stop using the engineering samples, once they have a finished, production ready product?
What OEMs took possesion of the chips almost a month ago, without first seeing a viable (production ready) board? Why would they take them knowing that it would be at least 2 months before they could put so much as a single chip to profitable use?
It must have been Dell, since everyone knows that Mikey has been wetting his pants over the opteron dual cores.
Then again IBM has already dropped 1.4 million$ to optimize a presler setup, so that it could perform as well as an opteron duallie system.

 

[ltcommander_data]

I noted that that contradiction as well. My take was that the majority of the volume shipments were the higher clocked 940, 950, and 955EE which do not have the stability problem. As well, Intel generally launchs their Extreme Editions earlier anyways as a technology preview. In addition, the stability is most likely limited to current 945 and 955 chipsets. Preslers are suppose to be able to work with the current motherboards out there right now, and it appears that the lower model Preslers are having trouble. That's why the motherboard manufacturers are hoping that they just need a BIOS update so that 945 and 955 board users can easily upgrade. The OEMs that took possesion of the higher clocked Preslers would plan to operate them on the new 975X that has just been released, which would avoid these stability problems.

 

[endyen]

Are you sure they work on 945/955 boards?
If they operate on any current boards, that would explain a lot. I was under the assumption that they required the new chipset.

 

[ltcommander_data]

Both Presler and Cedar Mill are supposed to be supported by chipsets that support the dual core Smithfield. Both the 945 and the 955 families should be supported in their entirety. Of course third party chipsets like nVidia is up to them. In any case the stability problem is most likely limited to the 945 and 955 chipset families for low-end Preslers only. High-end Preslers and Ceder Mills appear to operate fine. In my opinion, the problem is probably similar to what nVidia had with the 820 and the nForce4. Hopefully, the motherboard manufacturers will just need to implement a new BIOS update.

http://www.tomshardware.com/cpu/20051010/

"Fortunately, neither 65 nm product requires a new platform; both will work on the 945 and 955 chipset motherboards that are prevalent today."

The 975X was just a high-end chipset to support the 955EE and replace the 955X chipset. Mainstream processors will continue to be served by the 945 family. This would mean that the 975X has nothing to due with 965 family which is to be released with Conroe. Most likely the 975X will be refreshed as the 975XE or some other higher number at that time to support officially support DDR2 800 and probably unofficially support DDR2 1066. This will ensure they stay ahead of AMD's new M2 which supports DDR2 667 and possibly DDR2 800. The current 975X only officially supports DDR2 667 although DDR2 800 is unofficially supported.

 

[rettihSlluB]

Most likely the 975X will be refreshed as the 975XE or some other higher number at that time to support officially support DDR2 800 and probably unofficially support DDR2 1066. This will ensure they stay ahead of AMD's new M2 which supports DDR2 667 and possibly DDR2 800.

LOL :lol: :lol:
Well, ahead in memory speed, but not ahead in performance which really matters.

 

[ltcommander_data]

Well, it is quite interesting that Intel's architecture shows more benefit from increased bandwidth than from reduced latency.

Benchmarks
http://www.xbitlabs.com/articles/memory/display/ddr2-oc1ghz_8.html

Conclusions
http://www.xbitlabs.com/articles/memory/display/ddr2-oc1ghz_10.html

"Memory timings are the performance-determining factor with Athlon 64 systems, but they are not with Pentium 4 platforms. All of our tests indicate that systems with high-frequency memory deliver the maximum performance. This is especially conspicuous when the FSB is clocked at 1067MHz: the systems with DDR2-1067 and DDR2-800 SDRAM are always faster than the systems with DDR2-667 and DDR2-533 SDRAM irrespective of the selected timings."

I suppose if raw bandwidth benefits their architecture more than reduced latency, that's why Intel is just concentrating on increasing FSB and memory bandwidth for their Mobile and Desktop platforms rather than an integrated memory controller. Of course, 4-way and higher servers are another story.